kernel/watchdog.c: print traces for all cpus on lockup detection
authorAaron Tomlin <atomlin@redhat.com>
Mon, 23 Jun 2014 20:22:05 +0000 (13:22 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Mon, 23 Jun 2014 23:47:44 +0000 (16:47 -0700)
A 'softlockup' is defined as a bug that causes the kernel to loop in
kernel mode for more than a predefined period to time, without giving
other tasks a chance to run.

Currently, upon detection of this condition by the per-cpu watchdog
task, debug information (including a stack trace) is sent to the system
log.

On some occasions, we have observed that the "victim" rather than the
actual "culprit" (i.e.  the owner/holder of the contended resource) is
reported to the user.  Often this information has proven to be
insufficient to assist debugging efforts.

To avoid loss of useful debug information, for architectures which
support NMI, this patch makes it possible to improve soft lockup
reporting.  This is accomplished by issuing an NMI to each cpu to obtain
a stack trace.

If NMI is not supported we just revert back to the old method.  A sysctl
and boot-time parameter is available to toggle this feature.

[dzickus@redhat.com: add CONFIG_SMP in certain areas]
[akpm@linux-foundation.org: additional CONFIG_SMP=n optimisations]
[mq@suse.cz: fix warning]
Signed-off-by: Aaron Tomlin <atomlin@redhat.com>
Signed-off-by: Don Zickus <dzickus@redhat.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Mateusz Guzik <mguzik@redhat.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Jan Moskyto Matejka <mq@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Documentation/kernel-parameters.txt
Documentation/sysctl/kernel.txt
include/linux/nmi.h
kernel/sysctl.c
kernel/watchdog.c

index 884904975d0b98ec126f22b0a51b6b26569d0b99..c1b9aa8c5a52e807e6458d40d96d4f1bc91f107b 100644 (file)
@@ -3130,6 +3130,11 @@ bytes respectively. Such letter suffixes can also be entirely omitted.
                        [KNL] Should the soft-lockup detector generate panics.
                        Format: <integer>
 
+       softlockup_all_cpu_backtrace=
+                       [KNL] Should the soft-lockup detector generate
+                       backtraces on all cpus.
+                       Format: <integer>
+
        sonypi.*=       [HW] Sony Programmable I/O Control Device driver
                        See Documentation/laptops/sonypi.txt
 
index 708bb7f1b7e03cf89e7338ecc042aaaa6ba19333..c14374e71775f62ba1ed72a856a702cb1d819e47 100644 (file)
@@ -75,6 +75,7 @@ show up in /proc/sys/kernel:
 - shmall
 - shmmax                      [ sysv ipc ]
 - shmmni
+- softlockup_all_cpu_backtrace
 - stop-a                      [ SPARC only ]
 - sysrq                       ==> Documentation/sysrq.txt
 - sysctl_writes_strict
@@ -783,6 +784,22 @@ via the /proc/sys interface:
 
 ==============================================================
 
+softlockup_all_cpu_backtrace:
+
+This value controls the soft lockup detector thread's behavior
+when a soft lockup condition is detected as to whether or not
+to gather further debug information. If enabled, each cpu will
+be issued an NMI and instructed to capture stack trace.
+
+This feature is only applicable for architectures which support
+NMI.
+
+0: do nothing. This is the default behavior.
+
+1: on detection capture more debug information.
+
+==============================================================
+
 tainted:
 
 Non-zero if the kernel has been tainted.  Numeric values, which
index a17ab6398d7c275765fce939c8c125d564938484..447775ee2c4b0e51429af5c8e06276ac8a19c65d 100644 (file)
@@ -57,6 +57,7 @@ int hw_nmi_is_cpu_stuck(struct pt_regs *);
 u64 hw_nmi_get_sample_period(int watchdog_thresh);
 extern int watchdog_user_enabled;
 extern int watchdog_thresh;
+extern int sysctl_softlockup_all_cpu_backtrace;
 struct ctl_table;
 extern int proc_dowatchdog(struct ctl_table *, int ,
                           void __user *, size_t *, loff_t *);
index 075d1903138fd2ce2c212ab0e88beeed3c034c3a..75b22e22a72c1abd4865bf7c0313af5514970939 100644 (file)
@@ -860,6 +860,17 @@ static struct ctl_table kern_table[] = {
                .extra1         = &zero,
                .extra2         = &one,
        },
+#ifdef CONFIG_SMP
+       {
+               .procname       = "softlockup_all_cpu_backtrace",
+               .data           = &sysctl_softlockup_all_cpu_backtrace,
+               .maxlen         = sizeof(int),
+               .mode           = 0644,
+               .proc_handler   = proc_dointvec_minmax,
+               .extra1         = &zero,
+               .extra2         = &one,
+       },
+#endif /* CONFIG_SMP */
        {
                .procname       = "nmi_watchdog",
                .data           = &watchdog_user_enabled,
index 30e482240dae3552f2b0653075dc91251e4842ae..c3319bd1b0408c1f5822748a4d0b1567c799760d 100644 (file)
 
 int watchdog_user_enabled = 1;
 int __read_mostly watchdog_thresh = 10;
+#ifdef CONFIG_SMP
+int __read_mostly sysctl_softlockup_all_cpu_backtrace;
+#else
+#define sysctl_softlockup_all_cpu_backtrace 0
+#endif
+
 static int __read_mostly watchdog_running;
 static u64 __read_mostly sample_period;
 
@@ -47,6 +53,7 @@ static DEFINE_PER_CPU(bool, watchdog_nmi_touch);
 static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved);
 static DEFINE_PER_CPU(struct perf_event *, watchdog_ev);
 #endif
+static unsigned long soft_lockup_nmi_warn;
 
 /* boot commands */
 /*
@@ -95,6 +102,15 @@ static int __init nosoftlockup_setup(char *str)
 }
 __setup("nosoftlockup", nosoftlockup_setup);
 /*  */
+#ifdef CONFIG_SMP
+static int __init softlockup_all_cpu_backtrace_setup(char *str)
+{
+       sysctl_softlockup_all_cpu_backtrace =
+               !!simple_strtol(str, NULL, 0);
+       return 1;
+}
+__setup("softlockup_all_cpu_backtrace=", softlockup_all_cpu_backtrace_setup);
+#endif
 
 /*
  * Hard-lockup warnings should be triggered after just a few seconds. Soft-
@@ -271,6 +287,7 @@ static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
        unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts);
        struct pt_regs *regs = get_irq_regs();
        int duration;
+       int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace;
 
        /* kick the hardlockup detector */
        watchdog_interrupt_count();
@@ -317,6 +334,17 @@ static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
                if (__this_cpu_read(soft_watchdog_warn) == true)
                        return HRTIMER_RESTART;
 
+               if (softlockup_all_cpu_backtrace) {
+                       /* Prevent multiple soft-lockup reports if one cpu is already
+                        * engaged in dumping cpu back traces
+                        */
+                       if (test_and_set_bit(0, &soft_lockup_nmi_warn)) {
+                               /* Someone else will report us. Let's give up */
+                               __this_cpu_write(soft_watchdog_warn, true);
+                               return HRTIMER_RESTART;
+                       }
+               }
+
                printk(KERN_EMERG "BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n",
                        smp_processor_id(), duration,
                        current->comm, task_pid_nr(current));
@@ -327,6 +355,17 @@ static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
                else
                        dump_stack();
 
+               if (softlockup_all_cpu_backtrace) {
+                       /* Avoid generating two back traces for current
+                        * given that one is already made above
+                        */
+                       trigger_allbutself_cpu_backtrace();
+
+                       clear_bit(0, &soft_lockup_nmi_warn);
+                       /* Barrier to sync with other cpus */
+                       smp_mb__after_atomic();
+               }
+
                if (softlockup_panic)
                        panic("softlockup: hung tasks");
                __this_cpu_write(soft_watchdog_warn, true);